llm-app-development
Pass
Audited by Gen Agent Trust Hub on Oct 2, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill demonstrates patterns like RAG (Retrieval-Augmented Generation) and structured output analysis that ingest untrusted external data directly into model prompts.
- Ingestion points: Document chunks in the RAG pipeline (Pattern 1) and user-supplied product reviews (Pattern 2) in SKILL.md.
- Boundary markers: The code snippets use system instructions to constrain the model ("Answer questions using ONLY the provided context") and template literal delimiters to separate data from instructions.
- Capability inventory: The skill facilitates calls to the OpenAI API, vector database queries (PGVector), and tool execution loops (search, ticket creation).
- Sanitization: While the code snippets provide basic formatting, the skill's final checklist explicitly recommends adding prompt injection defenses such as input validation and output filtering.
Audit Metadata